Office backend follow-ups after 0.1.0 (from fetchkit integration) #11

These gaps came up while making markkit the default Office backend in fetchkit (everruns/fetchkit#183). They are ordered roughly by impact.

Format coverage

• PPTX support. fetchkit's deprecated in-crate converter handled it, so fetchkit loses PPTX by default until markkit supports it.
• XLSB support.
• Format::from_content_type maps application/vnd.ms-excel.sheet.binary.macroenabled.12 (XLSB) to Xlsx, because the string contains ms-excel.sheet. Detect it as XLSB instead.

Correctness

• A sheet with a single row renders as a header row with no data rows.
• Spreadsheets have no title: Extraction.title is always None for XLSX/XLS/ODS. The title could come from the document properties (docProps/core.xml or meta.xml).
• Output always ends with a trailing newline. Decide whether that is intended.

Safety and limits

• max_processing_secs is in the config but never enforced, so parsing has no time budget.
• The release profile uses panic = "abort", which makes the catch_unwind guard ineffective in markkit's own binaries.

API ergonomics for embedders

• Make Markkit Clone and Debug. fetchkit currently wraps it in an Arc and writes its own Debug.
• Make warnings structured (a code plus detail), not free-text code: detail strings.
• Give errors structured fields, for example ResourceLimit { size, limit }, so callers can map them without parsing strings.
• With default-features = false plus docx,spreadsheet, markkit still pulls in anyhow, regex, unicode-segmentation, mime, base64, serde_json and tokio. Make the ones the document path doesn't need optional.

Minor

• An XLSX file without _rels/.rels fails inside calamine. Real files always have it, so this only affects hand-built fixtures.